Steps in document processing

• To receive documents, you can use web upload, email, SFTP or REST API
• Support of unstructured data from PDFs or scanned documents
• Support of structured documents like CSV files or Excel

For images and scanned PDF documents, optical character recognition (OCR) is used to extract text directly from Excel tables. 

Determining the type of document ensures that the right data elements are extracted. The following are supported:

• Invoice
• Packing List
• Certificate of Origin
• and more

The documents are pre-classified by the user or via the upload API.

Extraction is performed using our Vision API including LLM models. The Vision API provides a self-assessment of the LLM’s performance regarding the quality of the extraction. Basic validations and data enhancements are directly implemented in MIC DOCFLOW.

Like all AI models, the Vision API can make mistakes. Therefore, the results should be verified by a human. MIC DOCFLOW enables users to compare the original document with the extracted data elements and make any necessary additions or corrections via a document editor.

In the final step, the data is made available in JSON format for further use in a subsequent module:

• either to be directly pushed to MIC modules, e.g., MIC-CUST^® workbuffers, MIC OCS, MIC INTRA…
• or to be downloaded for various use cases with 3^rd party software solutions